A Pipeline Extraction Algorithm for Forward-Looking Sonar Images Using the Self-Organizing Map

An autonomous underwater vehicle (AUV) can operate automatically with a small number of humans and requires few resources when compared to a remotely operated vehicle (ROV). The pipeline inspection with the use of an AUV can be significantly cheaper than those of an ROV. Successful deployment of AUVs relies on accurate pipeline detection and extraction that can automatically locate and track a pipeline direction in real time. We proposed to use forward-looking sonar images to locate and extract pipeline paths since sonar can easily image through turbulence and small particles, commonly found in underwater environments. However, sonar images are prone to disturbance from the environment. Thus, as a superior alternative, we developed a pipeline detection algorithm by considering that a pipeline has a piecewise linear shape. Next, the self-organizing map (SOM) is employed to join the labeled segments together to form a pipeline track to extract the pipeline path for the navigation of an AUV since SOM can map a 2-D image of pipeline into a 1-D line. In our experiment, data obtained from the Gulf of Thailand are used for the performance analysis, and we are able to extract and describe a pipeline direction with accuracies between 90% and 97%.